neuroeconomics - how neuroscience informs economics

8/8/2019 Neuroeconomics - How Neuroscience Informs Economics

1/56

Journal of Economic LiteratureVol. XLIII (March 2005), pp. 964

Neuroeconomics: How NeuroscienceCan Inform Economics

COLIN CAMERER, GEORGE LOEWENSTEIN , and DRAZEN PRELEC

Who knows what I want to do? Who knows what anyone wants to do? How can yoube sure about something like that? Isnt it all a question of brain chemistry, signalsgoing back and forth, electrical energy in the cortex? How do you know whethersomething is really what you want to do or just some kind of nerve impulse in thebrain. Some minor little activity takes place somewhere in this unimportant place inone of the brain hemispheres and suddenly I want to go to Montana or I dont wantto go to Montana. (White Noise, Don DeLillo)

9

Camerer: California Institute of Technology.Loewenstein: Carnegie Mellon University. Prelec:Massachusetts Institute of Technology. We thank partici-pants at the Russell Sage Foundation-sponsored confer-ence on Neurobehavioral Economics (May 1997) atCarnegie Mellon, the Princeton workshop on NeuralEconomics December 89, 2000, and the Arizona confer-ence in March 2001. This research was supported by NSFgrant SBR-9601236 and by the Center for Advanced Study in Behavioral Sciences, where the authors visited during199798. David Laibsons presentations have been partic-ularly helpful, as were comments and suggestions from theeditor and referees, and conversations and comments fromRalph Adolphs, John Allman, Warren Bickel, Greg Berns,Meghana Bhatt, Jonathan Cohen, Angus Deaton, JohnDickhaut, Paul Glimcher, Dave Grether, Ming Hsu, DavidLaibson, Danica Mijovic-Prelec, Read Montague, CharliePlott, Matthew Rabin, Antonio Rangel, Peter Shizgal,Steve Quartz, and Paul Zak. Albert Bollard, EstherHwang, and Karen Kerbs provided editorial assistance.

1. Introduction

In the last two decades, following almost acentury of separation, economics has begunto import insights from psychology.Behavioral economics is now a prominentfixture on the intellectual landscape and hasspawned applications to topics in economics,

such as finance, game theory, labor econom-ics, public finance, law, and macroeconomics(see Colin Camerer and George Loewenstein2004). Behavioral economics has mostly beeninformed by a branch of psychology calledbehavioral decision research, but othercognitive sciences are ripe for harvest. Someimportant insights will surely come from neu-roscience, either directly or because neuro-science will reshape what is believed aboutpsychology which in turn informs economics.

Neuroscience uses imaging of brain activity and other techniques to infer details abouthow the brain works. The brain is the ultimateblack box. The foundations of economictheory were constructed assuming that detailsabout the functioning of the brains black box would not be known. This pessimism wasexpressed by William Jevons in 1871:

I hesitate to say that men will ever have the meansof measuring directly the feelings of the humanheart. It is from the quantitative effects of thefeelings that we must estimate their comparativeamounts.

mr05_Article 1 3/28/05 3:25 PM Page 9


2/56

10 Journal of Economic Literature, Vol. XLIII (March 2005)

Since feelings were meant to predict behav-ior but could only be assessed from behavior,economists realized that, without direct

measurement, feelings were useless inter- vening constructs. In the 1940s, the con-cepts of ordinal utility and revealedpreference eliminated the superfluous inter-mediate step of positing immeasurable feel-ings. Revealed preference theory simply equates unobserved preferences withobserved choices. Circularity is avoided by assuming that people behave consistently, which makes the theory falsifiable; oncethey have revealed that they prefer A to B,people should not subsequently choose B

over A. Later extensionsdiscounted,expected, and subjective expected utility,and Bayesian updatingprovided similaras if tools which sidestepped psychologicaldetail. The as if approach made goodsense as long as the brain remained substan-tially a black box. The development of eco-nomics could not be held hostage toprogress in other human sciences.

But now neuroscience has provedJevonss pessimistic prediction wrong; thestudy of the brain and nervous system is

beginning to allow direct measurement of thoughts and feelings. These measurementsare, in turn, challenging our understandingof the relation between mind and action,leading to new theoretical constructs andcalling old ones into question. How can thenew findings of neuroscience, and the theo-ries they have spawned, inform an econom-ic theory that developed so impressively intheir absence?

In thinking about the ways that neuro-science can inform economics, it is useful to

distinguish two types of contributions, which we term incremental and radical approach-es. In the incremental approach, neuro-science adds variables to conventionalaccounts of decision making or suggests spe-cific functional forms to replace as ifassumptions that have never been well sup-ported empirically. For example, research onthe neurobiology of addiction suggests how

drug consumption limits pleasure fromfuture consumption of other goods (dynam-ic cross-partial effects in utility for commod-

ity bundles) and how environmental cuestrigger unpleasant craving and increasedemand. These effects can be approximatedby extending standard theory and thenapplying conventional tools (see DouglasBernheim and Antonio Rangel 2004; DavidLaibson 2001; Ted ODonoghue andMatthew Rabin 1997).

The radical approach involves turningback the hands of time and asking how eco-nomics might have evolved differently if ithad been informed from the start by insights

and findings now available from neuro-science. Neuroscience, we will argue, pointsto an entirely new set of constructs to under-lie economic decision making. The standardeconomic theory of constrained utility maxi-mization is most naturally interpreted eitheras the result of learning based on consump-tion experiences (which is of little help whenprices, income, and opportunity setschange), or careful deliberationa balanc-ing of the costs and benefits of differentoptionsas might characterize complex

decisions like planning for retirement, buy-ing a house, or hammering out a contract.Although economists may privately acknowl-edge that actual flesh-and-blood humanbeings often choose without much delibera-tion, the economic models as written invari-ably represent decisions in a deliberativeequilibrium, i.e., that are at a stage wherefurther deliberation, computation, reflec-tion, etc. would not by itself alter the agentschoice. The variables that enter into the for-mulation of the decision problemthe pref-

erences, information, and constraintsareprecisely the variables that should affect thedecision, if the person had unlimited timeand computing ability.

While not denying that deliberation is partof human decision making, neurosciencepoints out two generic inadequacies of thisapproachits inability to handle the crucialroles of automatic and emotional processing.



3/56

Camerer, Loewenstein, and Prelec: Neuroeconomics 11

First, much of the brain implementsautomatic processes, which are faster thanconscious deliberations and which occur

with little or no awareness or feeling of effort(John Bargh et al. 1996; Bargh and TanyaChartrand 1999; Walter Schneider andRichard Shiffrin 1977; Shiffrin andSchneider 1977). Because people have littleor no introspective access to these processes,or volitional control over them, and theseprocesses were evolved to solve problems of evolutionary importance rather than respectlogical dicta, the behavior these processesgenerate need not follow normative axiomsof inference and choice.

Second, our behavior is strongly influ-enced by finely tuned affective (emotion)systems whose basic design is common tohumans and many animals (Joseph LeDoux1996; Jaak Panksepp 1998; Edmund Rolls1999). These systems are essential for daily functioning, and when they are damaged orperturbed, by brain injury, stress, imbal-ances in neurotransmitters, or the heat of the moment, the logical-deliberative sys-temeven if completely intactcannotregulate behavior appropriately.

Human behavior thus requires a fluidinteraction between controlled and automat-ic processes, and between cognitive andaffective systems. However, many behaviorsthat emerge from this interplay are routine-ly and falsely interpreted as being the prod-uct of cognitive deliberation alone (George Wolford, Michael Miller, and MichaelGazzaniga 2000). These results (some of which are described below) suggest thatintrospective accounts of the basis for choiceshould be taken with a grain of salt. Because

automatic processes are designed to keepbehavior off-line and below consciousness, we have far more introspective access tocontrolled than to automatic processes.Since we see only the top of the automaticiceberg, we naturally tend to exaggerate theimportance of control.

Neuroscience findings and methods willundoubtedly play an increasingly prominent

1 The first meeting was held at Carnegie Mellon in1997. Later meetings were held in Arizona andPrinceton, in 2001, in Minnesota in 2002, and onMarthas Vineyard in 2003 and Kiawah Island in 2004.Occasional sessions devoted to this rapidly growing topicare now common at large annual meetings in botheconomics and neuroscience.

role in economics and other social sciences(e.g., law, Terrence Chorvat, Kevin McCabe,and Vernon Smith 2004). Indeed, a new area

of economics that has been branded neu-roeconomics has already formed the basisof numerous academic gatherings that havebrought neuroscientists and economiststogether (see also Paul Zak, in press).1Participating in the development of a sharedintellectual enterprise will help us ensurethat the neuroscience informs economicquestions we care about. Stimulated by theauthors own participation in a number of such meetings, our goal in this paper is todescribe what neuroscientists do and how

their discoveries and views of human behav-ior might inform economic analysis. In thenext section (2), we describe the diversity of tools that neuroscientists use. Section 3introduces a simplified account of how cog-nition and affect on the one hand, and auto-matic and controlled processes on the other, work separately, and interact. Section 4 dis-cusses general implications for economics.Section 5 goes into greater detail aboutimplications of neuroeconomics for four top-ics in economics: intertemporal choice, deci-

sion making under risk, game theory, andlabor-market discrimination. Through mostof the paper, our focus is largely on how neu-roscience can inform models of microfoun-dations of individual decision making.Section 6 discusses some broader macroimplications and concludes.

2. Neuroscience Methods

Scientific technologies are not just toolsscientists use to explore areas of interest.New tools also define new scientific fieldsand erase old boundaries. The telescope cre-ated astronomy by elevating the science from



4/56


pure cosmological speculation. The micro-scope made possible similar advances in biol-ogy. The same is true of economics. Its

boundaries have been constantly reshaped by tools such as mathematical, econometric, andsimulation methods. Likewise, the currentsurge of interest in neuroscience by psychol-ogists emerged largely from new methods,and the methods may productively blur theboundaries of economics and psychology.This section reviews some of these methods.

2.1 Brain Imaging

Brain imaging is currently the most popu-lar neuroscientific tool. Most brain imaging

involves a comparison of people performingdifferent tasksan experimental task anda control task. The difference betweenimages taken while subject is performing thetwo tasks provides a picture of regions of thebrain that are differentially activated by theexperimental task.

There are three basic imaging methods.The oldest, electro-encephalogram (or EEG)uses electrodes attached to the scalp to meas-ure electrical activity synchronized to stimu-lus events or behavioral responses (known as

Event Related Potentials or ERPs). LikeEEG, positron emission topography (PET)scanning is an old technique in the rapidly changing time-frame of neuroscience, but isstill a useful technique. PET measures bloodflow in the brain, which is a reasonable proxy for neural activity, since neural activity in aregion leads to increased blood flow to thatregion. The newest, and currently most pop-ular, imaging method is functional magneticresonance imaging (fMRI), which tracksblood flow in the brain using changes in mag-

netic properties due to blood oxygenation(the BOLD signal). Simultaneous directrecording of neural processing and fMRIresponses confirms that the BOLD signalreflects input to neurons and their processing(Nikos Logothetis et al. 2001).

Although fMRI is increasingly becomingthe method of choice, each method haspros and cons. EEG has excellent temporal

resolution (on the order of one millisecond)and is the only method used with humansthat directly monitors neural activity, as

opposed to, e.g., blood flow. But spatial reso-lution is poor and it can only measure activi-ty in the outer part of the brain. EEGresolution has, however, been improvingthrough the use of ever-increasing numbersof electrodes. Interpolation methods, and thecombined use of EEG and fMRI for measur-ing outer-brain signals and inner-brain sig-nals at the same time, promise to createstatistical methods which make reasonableinferences about activity throughout thebrain from EEG signals. For economics, a

major advantage of EEG is its relative unob-trusiveness and portability. Portability willeventually reach the point where it will bepossible to take unobtrusive measurementsfrom people as they go about their daily affairs. PET and fMRI provide better spatialresolution than EEG, but poorer temporalresolution because blood-flow to neurally active areas occurs with a stochastic lag froma few seconds (fMRI) to a minute (PET).

Brain imaging still provides only a crudesnapshot of brain activity. Neural processes

are thought to occur on a 0.1 millimeterscale in 100 milliseconds (msec), but thespatial and temporal resolution of a typicalscanner is only 3 millimeters and several sec-onds. Multiple trials per subject can be aver-aged to form composite images, but doing soconstrains experimental design. However,the technology has improved rapidly and willcontinue to improve. Hybrid techniques thatcombine the strengths of different methodsare particularly promising. Techniques forsimultaneously scanning multiple brains

(hyperscanning) have also been developed, which can be used to study multiple-brain-level differences in activity in games andmarkets (Read Montague et al. 2002).

2.2 Single-Neuron Measurement

Even the finest-grained brain imagingtechniques only measure activity of cir-cuits consisting of thousands of neurons. In



5/56


single neuron measurement, tiny electrodesare inserted into the brain, each measuring asingle neurons firing. As we discuss below,

single neuron measurement studies haveproduced some striking findings that, webelieve, are relevant to economics. A limita-tion of single neuron measurement is that,because insertion of the wires damagesneurons, it is largely restricted to animals.

Studying animals is informative abouthumans because many brain structures andfunctions of non-human mammals are simi-lar to those of humans (e.g., we are moregenetically similar to many species of mon-keys than those species are to other species).

Neuroscientists commonly divide the braininto crude regions that reflect a combinationof evolutionary development, functions, andphysiology. The most common, triune divi-sion draws a distinction between the reptil-ian brain, which is responsible for basicsurvival functions, such as breathing, sleep-ing, eating, the mammalian brain, whichencompasses neural units associated withsocial emotions, and the hominid brain, which is unique to humans and includesmuch of our oversized cortexthe thin, fold-

ed, layer covering the brain that is responsi-ble for such higher functions as language,consciousness and long-term planning (PaulMacLean 1990). Because single neuronmeasurement is largely restricted to nonhu-man animals, it has so far shed far more lighton the basic emotional and motivationalprocesses that humans share with othermammals than on higher-level processessuch as language and consciousness.

2.3 Electrical Brain Stimulation (EBS)

Electrical brain stimulation is anothermethod that is largely restricted to animals.In 1954, two psychologists (James Olds andPeter Milner 1954) discovered that rats would learn and execute novel behaviors if rewarded by brief pulses of electrical brainstimulation (EBS) to certain sites in thebrain. Rats (and many other vertebrates,including humans) will work hard for EBS.

For a big series of EBS pulses, rats will leapover hurdles, cross electrified grids, andforego their only daily opportunities to eat,

drink, or mate. Animals also trade EBS off against smaller rewards in a sensible fash-ione.g., they demand more EBS to foregofood when they are hungry. Unlike morenaturalistic rewards, EBS does not satiate.And electrical brain stimulation at specificsites often elicits behaviors such as eating,drinking (Joseph Mendelson 1967), or copu-lation (Anthony Caggiula and Bartley Hoebel 1966). Many abused drugs, such ascocaine, amphetamine, heroin, cannabis,and nicotine, lower the threshold at which

animals will lever-press for EBS (Roy Wise1996). Despite its obvious applications toeconomics, only a few studies have exploredsubstitutability of EBS and other reinforcers(Steven Hursh and B. H. Natelson; LeonardGreen and Howard Rachlin 1991; PeterShizgal 1999).

2.4 Psychopathology and Brain Damage inHumans

Chronic mental illnesses (e.g., schizophre-nia), developmental disorders (e.g., autism),

degenerative diseases of the nervous system,and accidents and strokes that damage local-ized brain regions help us understand howthe brain works (e.g., Antonio Damasio1994). When patients with known damage toan area X perform a special task more poor-ly than normal patients, and do other tasksequally well, one can infer that area X is usedto do the special task. Patients who haveundergone neurosurgical procedures such aslobotomy (used in the past to treat depres-sion) or radical bisection of the brain (an

extreme remedy for epilepsy, now rarely used) have also provided valuable data (see Walter Freeman and James Watts 1942;Gazzaniga and LeDoux 1978).

Finally, a relatively new method calledtranscranial magnetic stimulation (TMS)uses pulsed magnetic fields to temporarilydisrupt brain function in specific regions.The difference in cognitive and behavioral



6/56


7/56


low offers in an ultimatum bargaininggame. This means that even if rejecting alow offer is done because of an adapted

instinct to build up a reputation for tough-ness (in order to get more in the future),the circuitry that encodes this instinct clear-ly has an affective component, which is notpurely cognitive.

For neuroeconomists, knowing moreabout functional specialization, and howregions collaborate in different tasks, couldsubstitute familiar distinctions between cat-egories of economic behavior (sometimesestablished arbitrarily by suggestions whichbecome modeling conventions) with new

ones grounded in neural detail. For exam-ple, the insula activity noted by Sanfey et al.in bargaining is also present when subjectsplaying matrix games are asked to guess what other subjects think they will do (sec-ond-order beliefs) (see Meghana Bhatt andCamerer, in press). This suggests that maybethe subjects who receive low offers in theultimatum study are not disgusted, they aresimply evaluating a second-order belief about what proposers expect them to do, asan input into an emotional evaluation. This is

just a speculation, of course, but it showshow direct understanding of neural circuitry can inspire theorizing and the search fornew data.

In light of the long list of methodsreviewed earlier in this section, it is also worth emphasizing that neuroscience isntonly about brain imaging. Brain lesion stud-ies (as well as TMS) allow one to examinethe impact of disabling specific parts of thebrain. They have often provided clearer evi-dence on functionality of specific brain

regionseven with tiny sample sizes fromrare types of damagethan imaging meth-ods have. Single neuron measurement provides information not just about what partsof the brain light up, but about the specif-ic conditions that cause specific neurons tofire at accelerated or decelerated rates. Andelectrical brain stimulation, though it islargely off limits to studies involving

humans, provides a level of experimentalcontrol that economists conducting fieldresearch should envy.

As always, the proof is in the pudding.Here, our goal is not to review the many ways in which neuroscience will rapidly change economic theory, because that is not where we are yet. Our goal is to showcasesome key findings in neuroscience, and stim-ulate the readers curiosity about what thesefindings might mean for economics.

3. Basic Lessons from Neuroscience

Because most of these techniques involve

localization of brain activity, this can easily foster a misperception that neuroscience ismerely developing a geography of thebrain, a map of which brain bits do whatpart of the job. If that were indeed so, thenthere would be little reason for economiststo pay attention. In reality, however, neuro-science is beginning to use regional activity differences and other clues to elucidate theprinciples of brain organization and func-tioning, which in turn, is radically changingour understanding of how the brain works.

Our goal in this section is to highlight someof the findings from neuroscience that may prove relevant to economics.

3.1 A Two-Dimensional TheoreticalFramework

Our organizing theme, depicted in Table 1,emphasizes the two distinctions mentionedin the introduction, between controlled andautomatic processes, and between cognitionand affect.

3.1.1 Automatic and Controlled Processes

The distinction between automatic andcontrolled processes was first proposed by Schneider and Shiffrin (1977). Many othershave developed similar two-system modelssince then, with different labels: rule-basedand associative (Steven Sloman 1996);rational and experiential systems (LeeKirkpatrick and Seymour Epstein 1992);



8/56


TABLE 1T WO DIMENSIONS OF NEURAL FUNCTIONING

Cognitive AffectiveControlled Processes

serial effortful evoked deliberately I II good introspective

access

Automatic Processes parallel effortless reflexive III IV no introspective

access

2 Elaborate methods have been developed that maxi-mize the validity of such verbal protocols (see, e.g.,Herbert Simon).

reflective and reflexive (MatthewLieberman et al. 2002); deliberative andimplementive systems (Peter Gollwitzer,Kentaro Fujita, and Gabriele Oettingen2004); assessment and locomotion (ArieKruglanski et al. 2000), and type I and typeII processes (Daniel Kahneman and ShaneFrederick 2002).

Controlled processes, as described by thetwo rows of Table 1, are serial (they usestep-by-step logic or computations), tend tobe invoked deliberately by the agent whenher or she encounters a challenge or sur-prise (Reid Hastie 1984), and are oftenassociated with a subjective feeling of effort. People can typically provide a goodintrospective account of controlled process-es. Thus, if asked how they solved a mathproblem or choose a new car, they canoften recall the considerations and thesteps leading up to the choice.2 Standardtools of economics, such as decision treesand dynamic programming, can be viewedas stylized representations of controlledprocesses.

3 The brains ability to recover from environmentaldamage is also facilitated by a property called plasticity. Inone study that illustrates the power of plasticity, the opticnerves of ferrets (which are born when they are still at arelatively immature state of development when the brain isstill highly plastic) were disconnected at birth and recon-nected to the auditory cortex (the portion of the brain thatprocesses sound). The ferrets learned to see using audi-tory cortex, and some neurons in their auditory cortexactually took on the physical characteristics of neurons inthe visual cortex (Lauire von Melchner, Sarah Pallas, andMriganka Sur 2000).

Automatic processes are the opposite of controlled processes on each of thesedimensions they operate in parallel, arenot accessible to consciousness, and arerelatively effortless. Parallelism facilitatesrapid response, allows for massive multi-tasking, and gives the brain remarkablepower when it comes to certain types of tasks, such as visual identification.Parallelism also provides redundancy thatdecreases the brains vulnerability to injury. When neurons are progressively destroyedin a region, the consequences are typically gradual rather than sudden (gracefuldegradation).3 Connectionist neural network models formulated by cognitive psy-chologists (David Rumelhart and JamesMcClelland 1986) capture these featuresand have been applied to many domains,



9/56


Figure 1. The human brain with some economically relevant areas marked.

including commercial ones. Models of thistype have a very different structure fromthe systems of equations that economiststypically work with. Unlike systems of equations, they are black-box; it is hard tointuit what they are doing by looking atindividual parameters.

Because automatic processes are notaccessible to consciousness, people oftenhave surprisingly little introspective insight

into why automatic choices or judgments were made. A face is perceived as attrac-tive or a verbal remark as sarcastic auto-matically and effortlessly. It is only laterthat the controlled system may reflect onthe judgment and attempt to substantiate itlogically, and when it does, it often does sospuriously (e.g., Timothy Wilson, SamuelLindsey, and Tonya Schooler 2000).

Automatic and controlled processes canbe roughly distinguished by where they occur in the brain (Lieberman et al. 2002).Regions that support cognitive automaticactivity are concentrated in the back (occip-ital), top (parietal), and side (temporal)parts of the brain (see Figure 1). The amyg-dala, buried below the cortex, is responsiblefor many important automatic affectiveresponses, especially fear. Controlled

processes occur mainly in the front (orbitaland prefrontal) parts of the brain. The pre-frontal cortex (pFC) is sometimes called theexecutive region, because it draws inputsfrom almost all other regions, integratesthem to form near and long-term goals, andplans actions that take these goals intoaccount (Timothy Shallice and PaulBurgess 1996). The prefrontal area is the

Somato-sensory

Visualcortex Automatic

affect

Cognitivecontrolinterrupt

Brodmann10

L insula R insula



10/56


4 As David Laibson and Andrew Caplin, respectively,have aptly expressed it.

region that has grown the most in thecourse of human evolution and which,therefore, most sharply differentiates us

from our closest primate relatives (StephenManuck et al. 2003).Automatic processeswhether cognitive

or affectiveare the default mode of brainoperation. They whir along all the time,even when we dream, constituting most of the electro-chemical activity in the brain.Controlled processes occur at specialmoments when automatic processes becomeinterrupted, which happens when a per-son encounters unexpected events, experi-ences strong visceral states, or is presented

with some kind of explicit challenge in theform of a novel decision or other type of problem. To the degree that controlledprocesses are well described by economiccalculation but parallel processes are not,one could say that economics is about theinterrupt or override.4

3.1.2 Affective and Cognitive Processes

The second distinction, represented by the two columns of table 1, is betweenaffective and cognitive processes. Such a

distinction is pervasive in contemporary psychology (e.g., Robert Zajonc 1980, 1984,1998; Zajonc and Daniel McIntosh 1992)and neuroscience (Damasio 1994; LeDoux1996; Panksepp 1998), and has an historicallineage going back to the ancient Greeksand earlier (Plato described people as driv-ing a chariot drawn by two horses, reasonand passions).

The distinguishing features of affectiveprocessing are somewhat counterintuitive.Most people undoubtedly associate affect

with feeling states, and indeed most affectstates do produce feeling states when they reach a threshold level of intensity.However, most affect probably operatesbelow the threshold of conscious awareness

5 By this definition, neural processes that dont have valence are not affects. There are also neural processesthat produce actions that are not best defined as affects e.g., the reflexes that cause you to draw away from a hotobject or an electric shock.

(LeDoux 1996; Piotr Winkielman and KentBerridge 2004). As Rita Carter (1999) com-ments, the conscious appreciation of emo-

tion is looking more and more like onequite small, and sometimes inessential, ele-ment of a system of survival mechanismsthat mainly operateeven in adultsat anunconscious level.

For most affect researchers, the centralfeature of affect is not the feeling states asso-ciated with it, but its role in human motiva-tion. All affects have valencethey areeither positive or negative (though somecomplex emotions, such as bittersweet,can combine more basic emotions that have

opposing valences). Many also carry actiontendencies (Nico Frijda 1986; LeonardBerkowitz 1999)e.g., anger motivates us toaggress, pain to take steps to ease the pain,and fear to escape or in some cases tofreezeas well as diverse other effects onsensory perception, memory, preferences,and so on (see, e.g., Leda Cosmides andJohn Tooby 2004). Affective processes,according to Zajoncs (1998) definition, arethose that address go/no-go questionsthat motivate approach or avoidance behav-

ior. Cognitive processes, in contrast, arethose that answer true/false questions.5Though it is not essential to our overall argu-ment, our view is that cognition by itself can-not produce action; to influence behavior,the cognitive system must operate via theaffective system.

Affect, as we use the term, embodies notonly emotions such as anger, fear, and jeal-ousy, but also drive states such as hunger,thirst and sexual desire, and motivationalstates such as physical pain, discomfort

(e.g., nausea) and drug craving. Ross Buck(1999) refers to these latter influences asbiological affects, which he distinguishes



11/56


6 The researchers scanned the brains of subjects usingfMRI (functional magnetic resonance imaging) as they played a video game designed to produce a feeling of socialrejection. Subjects thought they were playing a game thatinvolved throwing a ball back and forth with two otherpeople, but in fact the computer controlled the two ani-mated figures that they saw on the screen. After a periodof three-way play, the two other players began to excludethe subjects by throwing the ball back and forth betweenthemselves. The social snub triggered neural activity in apart of the brain called the anterior cingulate cortex, whichalso processes physical pain, and the insula, which is activeduring physical and social discomfort.

from the more traditional social affects.Affect, thus, coincides closely with the his-torical concept of the passions. Although

emotions such as anger and fear mightseem qualitatively different than the bio-logical affects, they have more in commonthat might be supposed. Thus, a recentstudy showed that hurt feelings activatedthe same brain regions activated by brokenbones or other physical injuries (NaomiEisenberger et al. 2003).6

3.1.3 The Quadrants in Action: AnIllustration

These two dimensions, in combination,

define four quadrants (labeled I to IV inTable 1). Quadrant I is in charge when youdeliberate about whether to refinance yourhouse, poring over present-value calcula-tions; quadrant II is undoubtedly the rarestin pure form. It is used by method actors who imagine previous emotional experi-ences so as to actually experience thoseemotions during a performance; quadrantIII governs the movement of your hand as you return serve; and quadrant IV makes you jump when somebody says Boo!

Most behavior results from the interac-tion of all four quadrants. A natural instinctof economists trained in parsimoniousmodeling is to think that cognition is typi-cally controlled, and affect is automatic, sothere are really only two dimensions(quadrants I and IV) rather than four. Buta lot of cognitive processing is automatic as welle.g., visual perception or language

7 For example, when people are shown unpleasant pho-tographs and told to interpret the photos so that they dontexperience negative emotions (e.g., imagine a picture of women crying as having been taken at a wedding), there isless activity in the insula and amygdala, emotional areas,and in medial orbitofrontal cortex, and area which theinsula projects to (Kevin Ochsner et al., in press).

processing. Research on emotional regu-lation shows many ways that cognitioninfluences emotion, which implies the

capacity for controlling emotion.7

The four-quadrant model is just a way toremind readers that the cognitiveaffectiveand controlledautomatic dimensions arenot perfectly correlated, and to provide abroad view to guide exploratory research.For some purposes, reducing the twodimensions to one, or the four quadrants totwo, will certainly be useful. Furthermore,noting all four cells is not a claim that all areequally important. It is just a suggestion thatleaving out one of the combinations would

lead to a model which is incomplete forsome purposes.Consider what happens when a party host

approaches you with a plate of sushi.Quadrant III: Your first task is to figure

out what is on the plate. The occipital cor-tex in the back of the brain is the first on thescene, drawing in signals from your eyes via your optic nerves. It decodes the sushi intoprimitive patterns such as lines and cornersthen uses a cascading process to discernlarger shapes (Stephen Kosslyn 1994).

Further downstream, in the inferior tempo-ral visual cortex (ITVC), this informationbecomes integrated with stored representa-tions of objects, which permits you to rec-ognize the objects on the plate as sushi. Thislatter process is extraordinarily complicated(and has proved difficult for artificial intelli-gence researchers to recreate in computers)because objects can take so many forms,orientations, and sizes.

Quadrant IV: This is where affect entersthe picture. Neurons in the inferior tempo-

ral visual cortex are sensitive only to the



12/56


identity of an object; they dont tell you whether it will taste good. Outputs of theinferior temporal visual cortex as well as out-

puts from other sensory systems feed intothe orbitofrontal cortex to determine thereward value of the recognized object.This is a highly particular representation. Ineconomic terms, what is represented is nei-ther pure information (i.e., that this is sushi)nor pure utility (i.e., that it is something Ilike) but rather a fusion of informationandutility. It is as if certain neurons in theorbitofrontal cortex are saying this is sushiand I want it.

The reward value of sushi depends in

turn on many factors. First, there is yourpersonal history with sushi. If you got sickon sushi in the past, you will have an uncon-scious and automatic aversion to it (tasteaversion conditioning). The amygdalaseems to play a critical role in this kind of long-term learning (LeDoux 1996).Second, the reward value of the sushi willdepend on your current level of hunger;people can eat almost anythinggrass,bugs, human fleshif they are hungry enough. The orbitofrontal cortex and a sub-

cortical region called the hypothalamus aresensitive to your level of hunger (Rolls1999). Neurons in these regions fire morerapidly at the sight or taste of food when you are hungry, and fire less rapidly when you are not hungry.

Quadrants I and II: Processing oftenends before quadrants I and II go to work.If you are hungry, and like sushi, yourmotor cortex will guide your arm to reachfor the sushi and eat it, drawing on auto-matic quadrant III (reaching) and IV

(taste and enjoyment) processes. Undersome circumstances, however, higher levelprocessing may enter the picture. If yousaw a recent documentary on the risks of eating raw fish, you may recoil; or if youdislike sushi but anticipate disappointmentin the eyes of your proud host who madethe sushi herself, youll eat it anyway (orpick it up and discreetly hide it in a napkin

8 Paul Romer (2000) uses the example of peanut tastesas an illustration of how understanding the cause of revealed preference matters. One person loves the taste of peanuts, but is allergic to them and knows that the conse-quences of eating would be disastrous. When she is hun-gry, her visceral system motivates her to eat peanuts, buther deliberative system, with its ability to consider delayedconsequences, inhibits her from eating them. The otherperson developed a taste aversion to peanuts many yearsago, as a result of having gotten sick right after eatingthem. She knows at a cognitive level that the peanuts werenot the cause of her sickness, but her visceral system over-rules cognitive awareness. Revealed preference theory would stop at the conclusion that both women have disu-tility for peanuts. But the fact that the mechanisms under-lying their preferences are different leads to predictabledifferences in other kinds of behavior. For example, thetaste-averse woman will have a higher price elasticity (shell eat peanuts if paid enough) and she will learn toenjoy peanuts after eating them a few times (her taste-aversion can be extinguished). The allergic woman willalso like the smell of peanuts, which the taste-averter wont. Treatments also differ: cognitive therapies for treat-ing harmless phobias and taste-aversion train people to usetheir conscious quadrant I processing to overrule visceralquadrant IV impulses, whereas treatment of the allergicpeanut-avoider will concentrate on a medical cure.

when she turns to serve other guests).8These explicit thoughts involve anticipatedfeelings (your own and the hosts) and

draw on explicit memories from a part of the brain called the hippocampus (see fig-ure 1), inputs from the affective system(sometimes referred to as the limbic sys-tem), and anticipation (planning) fromthe prefrontal cortex.

Because standard economics is bestdescribed by the controlled, cognitive,processes of quadrant 1, in the remainder of this section we focus on the other half of each dichotomyautomatic and affectiveprocessesproviding further details of their

functioning.3.2 Automatic Processes

Here, we review some key principles of neural functioning that characterize auto-matic processes. Our short-list includes: parallelism, specialization, and coordina- tion. Unpacking this a bit, we would say that:(1) much of the brains processing involvesprocesses that unfold in parallel and are not



13/56


accessible to consciousness; (2) the brainutilizes multiple systems specialized to per-form specific functions; and (3) it figures out

how to use existing specialized systems toaccomplish new tasks efficiently, whateverfunctions they originally evolved to perform.

3.2.1 Parallelism

The brain performs a huge number of dif-ferent computations in parallel. Because of themassively interconnected network architec-ture of neural systems, computations done inone part of the brain have the potential toinfluence any other computation, even whenthere is no logical connection between the two.

Recent psychological research on automat-ic processing provides many striking examplesof such spurious interactions. In one particu-larly clever study, Gary Wells and RichardPetty (1980) had subjects listen to an editori-al delivered over headphones while shakingtheir head either updown, or rightleft (subjects were led to believe that shaking was alegitimate part of the product test). Those who were instructed to shake updownreported agreeing with the editorial morethan those instructed to shake leftright, pre-

sumably because in our culture updownhead movement is associated with an attitudeof acceptance, and leftright with an attitudeof rejection. A similar impact on preference was also observed in a study in which subjects were asked to evaluate cartoons while holdinga pen either clamped horizontally betweentheir teeth or grasping it with pursed lips as if puffing on a cigarette (Fritz Strack, LeonardMartin, and Sabine Stepper 1988). The hori-zontal clamp forces the mouth into a smile, which enhances ratings, while puffing

forces the mouth into a pursed expression, which lowers ratings. What the brain seems tobe doing, in these examples, is seeking aglobal equilibrium that would reconcile theforced action (e.g., forced smile) with theresponse and the perceived attributes of theevaluated object.

Given that people are capable of delibera-tion, why doesnt quadrant I thinking correct

the automatic activity produced by quadrantIII and IV processes when they are wrong?Indeed sometimes it does: pilots learn to

trust their instrument panels, even whenthey conflict with strong sensory intuitionsabout where their plane is headed, but suchcognitive override is more the exceptionthan the rule. To override automaticprocesses, quadrant I has to (a) recognizethat an initial impression is wrong (whichrequires self-awareness about behavior inthe other quadrants), and then (b) deliber-ately correct that impression. But whensense making works outside of consciousnessit will not generate alarm bells to trigger the

recognition required in (a). This is surely thecase in the Epley and Gilovich studies. Even when the external influence is obvious andinappropriate, or the subject is warnedahead of time, the deliberation required tocorrect the first impression is hard work,competing for mental resources and atten-tion with all the other work that needs to bedone at the same moment (Daniel Gilbert2002). The struggle between rapid uncon-scious pattern-detection processes and theirslow, effortful modulation by deliberation is

not a fair contest; so automatic impressions will influence behavior much of the time.

3.2.2 Specialization

Neurons in different parts of the brain havedifferent shapes and structures, different func-tional properties, and operate in coordinationas systems that are functionally specialized.Progress in neuroscience often involves trac-ing well-known psychological functions to cir-cumscribed brain areas. For example, Brocasand Wernickes areas are involved, respective-

ly, in the production and comprehension of language. Patients with Wernicke damage can-not understand spoken words. While they canproduce words themselves, they cant monitortheir own speech, which results in sentencesof correctly articulated words strung togetherinto unintelligible gibberish. People with dam-age to Brocas area, in contrast, can understand what is said to them and typically know what



14/56


9 Retrograde amnesia is the more familiar kind, in which people forget their past but can form new memo-ries. The 2000 movie Memento drew an accurate por-trait of anterograde amnesia andinterestingly foreconomists its vulnerability to exploitation by people who know you are forgetting what they did to you.

they want to say, but have difficulty articulat-ing it, sometimes to the point of not being ableto generate any words at all.

Beyond uncovering the nature of thesespecialized systems, neuroscience has led tothe discovery of new functional systems,some of which are quite surprising. Forexample, surgeons conducting brain surgery on an epileptic patient discovered a smallregion of her brain which, when stimulated,caused her to laugh (Itzhak Fried 1998),hinting at the existence of a humor system.Neuroscientists have also located an area inthe temporal lobe that, when stimulatedelectrically, produces intense religious feel-

ingse.g., the sense of a holy presence oreven explicit visions of God or Christ, evenin otherwise unreligious people (MichaelPersinger and Faye Healey 2002).

More generally, neuroscience has begunto change our classification of functionalprocesses in the brain, in some cases identi-fying distinct brain processes that serve thesame function, and in others drawing con-nections between processes that have beencommonly viewed as distinct.

An example of the former is memory.

Studies of patients with localized brain dam-age have confirmed the existence of distinctmemory systems, which can be selectively knocked out. Anterograde amnesiacs, forexample, are commonly able to recall infor-mation acquired before injury, and they areable to acquire implicit information, includ-ing perceptual-motor skills (like the ability toread text in a mirror), but they are unable torecall new explicit information for more thana minute.9 Anterograde amnesiacs can alsoacquire new emotional associations without

the explicit memories necessary to makesense of them. In one famous example, a doc-tor introduced himself with a tack concealed

in his hand that pricked a patient when they shook hands. Later the doctor returned and,although he did not remember the doctor, he

nevertheless responded negatively to thedoctors arrival and refused to shake his hand.An example of the latter involves the expe-

rience of fear, and recognition of fear in oth-ers. Experiencing an emotion andrecognizing the emotion when displayedintuitively seems to involve entirely differentprocesses. However, recent findings suggestthat this is not the case. Numerous studieshave implicated the amygdalaa smallorgan in the brain that is also intimately linked to the sense of smellto fear pro-

cessing. Lesions to the amygdala of rats andother animals disrupt or even eliminate theanimals fear-responses. Humans with strokedamage in the amygdala show similardeficits in reacting to threatening stimuli.

The same damage that disrupts fearresponses in humans also cripples a personsability to recognize facial expressions of fearin others, and to represent such expressionsin pictures. Figure 2 shows how a patient with amygdala damage expressed differentemotions when asked to draw them in the

form of facial expressions. The patient wasable to render most emotions with remark-able artistic talent. But when it came todrawing an expression of fear she felt clue-less. She didnt even try to draw an adultface; instead she drew a picture of a crawlinginfant looking apprehensive.

The finding that people who dont experi-ence fear also cant recognize it or represent itpictorially, suggests that two phenomena thathad been viewed as distinctexperiencingand representing fearin fact have important

commonalities. Beyond this, they point to theintriguing notion that to recognize an emotionin others, one needs to be able to experienceit oneself (see Alvin Goldman 2003).

The idea that people have specialized sys-tems that are invoked in specific situationscould have dramatic consequences for eco-nomics. The standard model of economicbehavior assumes that people have a unitary



15/56


Figure 2. Representations of different emotions drawn by a patient with amygdala damage(Adolphs et al. 1995).

set of preferences which they seek to satisfy,and economists often criticize psychology forlacking such a unified perspective. The exis-

tence of such selectivelyinvoked, special-ized, systems, however, raises the questionof whether a unified account of behavioris likely to do a very good job of capturingthe complexities of human behavior. AsJonathan Cohen (personal communication)aptly expressed it at a recent conference onneuroeconomics, Economics has one theo-ry, psychologists manyperhaps because

brains have different systems that they use tosolve different problems.

3.2.3 CoordinationIn a process that is not well understood,

the brain figures out how to do the tasks it isassigned efficiently, using the specialized sys-tems it has at its disposal. When the brain isconfronted with a new problem it initially draws heavily on diverse regions, including,often, the prefrontal cortex (where controlled



16/56


Figure 3. Regions of brain activation when first playing Tetris (left) and after several weeks of practice (right) (Haier et al. 1992).

processes are concentrated). But over time,activity becomes more streamlined, concen-trating in regions that specialized in process-ing relevant to the task. In one study

(Richard Haier et al. 1992), subjects brains were imaged at different points in time asthey gained experience with the computergame Tetris, which requires rapid handeyecoordination and spatial reasoning. Whensubjects began playing, they were highly aroused and many parts of the brain wereactive (Figure 3, left panel). However, as they got better at the game, overall blood flow tothe brain decreased markedly, and activity became localized in only a few brain regions(Figure 3, right panel).

Much as an economy ideally adjusts to theintroduction of a new product by gradually shifting production to the firms that can pro-duce the best goods most cheaply, with expe-rience at a task or problem, the brain seemsto gradually shift processing toward brainregions and specialized systems that cansolve problems automatically and efficiently with low effort.

Andrew Lo and Dmitry Repin (2002)observe a similar result in a remarkablestudy of professional foreign-exchange andderivatives traders who were wired for psy-

chophysiological measurements while they traded. Less-experienced traders showedsignificant physiological reactions to abouthalf of the market events (e.g., trend rever-sals). More experienced traders reactedmuch less to the same events. Years of automatizing apparently enabled seasonaltraders to react calmly to dramatic eventsthat send a novice trader on an emotionalroller coaster.

Given the severe limitations of controlledprocesses, the brain is constantly in the

process of automating the processing of tasksi.e., executing them using automaticrather than controlled processes. Indeed,one of the hallmarks of expertise in an area isthe use of automatic processes such as visualimagery and categorization. In one now-famous study, Fernand Gobet and Simon(1996) tested memory for configurations of chess pieces positioned on a chessboard.



17/56


They found that expert chess players wereable to store the positions of players almostinstantlybut only if they were in positions

corresponding to a plausible game. For ran-domly arranged chess pieces, the experts were not much better than novices. Furtherinvestigations have found that chess grand-masters store roughly 10,000 different possi-ble board setups in memory which they canrecognize almost instantly and respond to.More recent research in decision makingsuggests that this is a more general phenom-enon, because decision making often takesthe form of pattern matching rather than of an explicit weighing of costs and benefits

(e.g., Robyn Leboeuf 2002; Doug Medinand Max Bazerman 1999).In some cases, repeated use of particular

specialized systems can produce physically recognizable changes. Studies have foundthat, for example, violinists who finger violinstrings with their left hand show enlargeddevelopment of cortical regions which corre-spond to fingers on the left hand (ThomasElbert et al. 1995), and the brain regionsresponsible for navigation and spatial mem-ory (the hippocampus) of London taxi driv-

ers are larger than comparable areas innon-taxi drivers (Eleanor Maguire et al.2000). While the exact causality is difficult todetermine (perhaps preexisting differencesin size of brain region affect peoples talentsand occupational choices), a causal connec-tion in the posited direction has beendemonstrated in songbirds, whose brainsshow observable differences as a function of whether or not they have been exposed tothe song that is characteristic of their species(Carol Whaling et al. 1997).

3.2.4 The Winner Take All Nature of Neural Processing

Another feature that is reminiscent of theoperation of an economy is the winner-take-all nature of neural information processing(M. James Nichols and Bill Newsome 2002). While there is a lot that we do not knowabout the way in which the huge amount of

10 Alternatively, one could imagine that perceptions orbeliefs aggregate all relevant neural information (which would also be more in tune with Bayesian updating). So faras we can tell, the brain can use both principles of aggre-gation: when different neural populations carry similarinformation, then the overall perceptual judgment is anaverage of all information; when they carry very differentinformation, then the overall judgment follows the win-ner-take-all rule (Nichols and Newsome 2002). For a sim-ple model of a neural network that exhibits theseproperties, see Richard Hahnloser et al. 2000.

neural activity is distilled into a categoricalpercept, or a decision, we do know that thebrain doesnt invariably integrate (i.e., aver-

age) over the signals carried by individualneurons. In particular, when two distinctgroups of neurons convey different informa-tion about the external world, the resultingperceptual judgment often adopts the infor-mation of one neuronal group and entirely suppresses the information carried by theother. This is referred to as a winner-take-all principle of neural signal-extraction. As aresult, many brain processes are fundamen-tally categorical, yielding well-defined per-ceptions and thoughts even if the incoming

information is highly ambiguous.10

Theadvantages of this principle are clear if theultimate job of the brain is to initiate a dis-crete action, or categorize an object as one of several discrete types. The disadvantage isthat updating of beliefs in response to newinformation can procede in fits and starts, with beliefs remaining static as long as thenew information does not lead to recatego-rization, but changing abruptly and dramati-cally when the accumulation of evidenceresults in a change of categorization (see

Sendhil Mullainathan 2002).3.3 Affective Processes

The way the brain evolved is critical tounderstanding human behavior. In many domains, such as eating, drinking, sex, andeven drug use, human behavior resemblesthat of our close mammalian relatives, whichis not surprising because we share many of the neural mechanisms that are largely responsible for these behaviors. Many of the



18/56


processes that occur in these systems areaffective rather than cognitive; they aredirectly concerned with motivation. This

might not matter for economics were it notfor the fact that the principles that guide theaffective systemthe way that it operatesis so much at variance with the standardeconomic account of behavior.

3.3.1 Primacy of Affect

In contrast to the intuitive view of humanbehavior as driven by deliberations aboutcosts and benefits, it does not do a terribleinjustice to the field of psychology to say thata growing consensus has developed around

the view that affect is primary in the sensethat it is first on the scene and plays a dom-inant role in behavior. Indeed, as we discussbelow (section 3.4.2), the conscious brainoften erroneously interprets behavior thatemerges from automatic, affective, processesas the outcome of cognitive deliberations.

In a series of seminal papers, Zajonc (1980,1984, 1998) presented the results of studies which showed, first, that people can oftenidentify their affective reaction to some-thingwhether they like it or notmore

rapidly than they can even say what it is, and,second, that affective reactions to things canbe dissociated from memory for details of those things, with the former often being bet-ter. For example, we often remember whether we liked or disliked a particular per-son, book or movie, without being able toremember any other details (Bargh 1984).Subsequent research in social psychology takes Zajoncs initial research a step furtherby showing that the human brain affectively tags virtually all objects and concepts, and

that these affective tags are brought to mindeffortlessly and automatically when thoseobjects and concepts are evoked (e.g., RussellFazio et al. 1986; Anthony Greenwald, MarkKlinger, and Thomas Liu 1989; Greenwald1992; Bargh, Shelly Chaiken, PaulaRaymond, and Charles Hymes 1996; Jan DeHouwer, Dirk Hermans, and Paul Eelen1998; David Houston and Fazio 1989)

11 The amygdala has excellent properties for perform-ing such a function. Recent research in which amygdalaeof subjects were scanned while threatening stimuli wereflashed at various positions in the visual field found thatamygdala activation is just as rapid and just as pronounced when such stimuli are presented outside as when they arepresented inside the region of conscious awareness (AdamAnderson et al. 2003).

LeDoux and his colleagues (summarized inLeDoux 1996) have arrived at similar conclu-sions about the primacy of affect using very

different research methods and subjects.Based on studies conducted on rats, they dis-covered that there are direct neural projec-tions from the sensory thalamus (whichperforms crude signal-processing) to theamygdala (which is widely believed to play acritical role in the processing of affectivestimuli) that are not channeled through theneocortex. As a result, animals can have anaffective reaction to stimuli before their cor-tex has had the chance to perform morerefined processing of the stimulithey are lit-

erally afraid before knowing whether they should be. Such immediate affective respons-es provide organisms with a fast but crudeassessment of the behavioral options they face which makes it possible to take rapid action.They also provide a mechanism for interrupt-ing and refocusing attention (Simon 1967),typically shifting processing from automatic tocontrolled processing. As Jorge Armony et al.(1995) note, A threatening signal, such asone indicating the presence of a predator,arising from outside of the focus of attention

will have a reduced representation in the cor-tex. Thus, if the amygdala relied solely on thecortical pathway to receive sensory informa-tion, it would not be capable of processing,and responding to, those danger signals thatare not within the focus of attention (see also,Armony et al. 1997; Gavin DeBecker 1997).11

A similar pattern to the one LeDouxobserved in rats can also be seen in humans.Gilbert and Michael Gill (2000) propose thatpeople are momentary realists who trusttheir immediate emotional reactions and

only correct them through a comparatively laborious cognitive process. If the car behind



19/56


12 The mechanisms that monitor the bodys state canbe exquisitely complex, and can include both internal andexternal cues. In the case of nutritional regulation, forexample, internal cues include gastric distension (JamesGibbs et al. 1981), receptors sensitive to the chemicalcomposition of the food draining from the stomach D.Greenberg, Smith and Gibbs (1990); Arthur L.Campfield and Smith 1990). External cues include timeof day, estimated time till the next feeding, and the sightor smell of food.

you honks after your light turns green, youare likely to respond with immediate anger,followed, perhaps, by a sheepish acknowl-

edgment that maybe the honking personbehind you had a point, as you were dis-tracted when the light turned green.

It is important to note that while emotionsmay be fleeting, they can have a large eco-nomic impact if they create irreversible rashdecisions (as in crimes of passion). Evenemotions which could be transitory, such asembarrassment, can have long-run effects if they are kept alive by memory and socialreminders. For example, Dora Costa andMatthew Kahn (2004) found that deserters

in the Union Army during the Civil War, who were allowed to return to their hometowns without explicit penalty, often moved away because of the shame and social ostracism of being known to their neighbors as a deserter.

3.3.2 Homeostasis

To understand how the affective systemoperates, one needs to recall that humans didnot evolve to be happy, but to survive andreproduce. An important process by whichthe body attempts to achieve these goals is

called homeostasis. Homeostasis involvesdetectors that monitor when a system departsfrom a set-point,12 and mechanisms thatrestore equilibrium when such departuresare detected. Someindeed mostof thesemechanisms do not involve deliberate action.Thus, when the core body temperature fallsbelow the 98.6 F set-point, blood tends tobe withdrawn from extremities, and when itrises above the set-point one begins to sweat.But other processes do involve deliberateactione.g., putting on ones jacket when

cold or turning on the air conditioner whenhot. The brain motivates one to take suchactions using both a carrot and a stick.

The stick reflects the fact that departingfrom a set-point usually feels bade.g., itfeels bad to be either excessively hot orcoldand this negative feeling motivatesone to take actions that move one backtoward the set-point. The carrot is a processcalled alliesthesia (Michael Cabanac1979) whereby actions that move onetoward the set-point tend to feel pleasura-ble. When the body temperature falls below98.6, for example, almost anything thatraises body temperature (such as placing

ones hand in warm water) feels good, andconversely when the body temperature iselevated almost anything that lowers body temperature feels good.

As economists, we are used to thinking of preferences as the starting point for humanbehavior and behavior as the ending point. Aneuroscience perspective, in contrast, viewsexplicit behavior as only one of many mech-anisms that the brain uses to maintain home-ostasis, and preferences as transient state variables that ensure survival and reproduc-

tion. The traditional economic account of behavior, which assumes that humans act soas to maximally satisfy their preferences,starts in the middle (or perhaps even towardthe end) of the neuroscience account.Rather than viewing pleasure as the goal of human behavior, a more realistic account would view pleasure as a homeostatic cuean informational signal.13,14

13 Even deliberative behavior generated by quadrant Iis typically organized in a fashion that resembles home-ostasis (Miller, Eugene Gallanter, and Karl Pribram 1960;Loewenstein 1999). Rather than simply maximizing pref-erences in a limitless fashion, people set goals for them-selves, monitor their progress toward those goals, andadjust behavior when they fall short of their goals.

14 Of course, even the neuroscience account begins, insome sense, in the middle. A more complete understand-ing of behavior would also ask how these different mecha-nisms evolved over time. Since evolution selects for genesthat survive and reproduce, the result of evolution isunlikely to maximize pleasure or minimize pain. (See Gary Becker and Luis Rayo 2004.)



20/56


An important feature of many homeosta-tic systems is that they are highly attuned tochanges in stimuli rather than their levels. A

dramatic demonstration of such sensitivity to change came from single-neuron studiesof monkeys responding to juice rewards(see Wolfram Schultz and Anthony Dickinson 2000). These studies measuredthe firing of dopamine neurons in the ani-mals ventral striatum, which is known toplay a powerful role in motivation andaction. In their paradigm, a tone was sound-ed, and two seconds later a juice reward wassquirted into the monkeys mouth. Initially,the neurons did not fire until the juice was

delivered. Once the animal learned that thetone forecasted the arrival of juice two sec-onds later, however, the same neurons firedat the sound of the tone, but did not fire when the juice reward arrived. These neu-rons were not responding to reward, or itsabsence . . . they were responding to devia-tions from expectations. (They are some-times called prediction neurons.) Whenthe juice was expected from the tone, but was not delivered, the neurons fired at a very low rate, as if expressing disappoint-

ment. The same pattern can be observed ata behavioral level in animals, who will workharder (temporarily) when a reinforcementis suddenly increased and go on strike when reinforcement falls.15 Neural sensitiv-ity to change is probably important inexplaining why the evaluation of risky gam-bles depends on a reference point which

15 As Rolls (1999) writes, We are sensitive to someextent not just to the absolute level of reinforcement beingreceived, but also to the change in the rate or magnitudeof reinforcers being received. This is well shown by thephenomena of positive and negative contrast effects withrewards. Positive contrast occurs when the magnitude of areward is increased. An animal will work very much hard-er for a period (perhaps lasting for minutes or longer) inthis situation, before gradually reverting to a rate close tothat at which the animal was working for the small rein-forcement. A comparable contrast effect is seen when thereward magnitude (or rate at which rewards are obtained)is reducedthere is a negative overshoot in the rate of working for a time.

encodes whether an outcome is a gain or aloss (see section 5), why self-reported hap-piness (and behavioral indicators like sui-

cide) depend on changes in income and wealth, rather than levels (Andrew Oswald1997), and why violations of expectationstrigger powerful emotional responses(George Mandler 1982).

3.4 Interactions between the Systems

Behavior emerges from a continuousinterplay between neural systems support-ing activity within each of the four quad-rants. Three aspects of this interaction bearspecial emphasis, which we labeled collab-

oration, competition, and sense-mak-ing. Collaboration captures the insight thatdecision making, which is to say rationalityin the broad, nontechnical sense of the word, is not a matter of shifting decision-making authority from quadrants II, III, andIV toward the deliberative, nonaffectivequadrant I, but more a matter of maintain-ing proper collaboration in activity across allfour quadrants. If quadrant I tries to do the job alone, it will often fail.16 Competitionreflects the fact that different processes

most notably affective and cognitiveoftendrive behavior in conflicting directions andcompete for control of behavior. Sense-making refers to how we make sense of suchcollaboration and competitionhow wemake sense of our behavior. While behavioris, in fact, determined by the interaction of all four quadrants, conscious reflection onour behavior, and articulating reasons for it,is basically quadrant I trying to make senseof that interaction. And, not surprisingly,

16 Roy Baumeister, Todd Heatherton, and Dianne Tice(1994) review studies which suggest that excessively highincentives can result in supra-optimal levels of motivationthat have perverse effects on performance (known as theYerkesDodson law). Beyond documenting such chokingunder pressure, the review research showing that it oftenoccurs because it causes people to utilize controlledprocesses for performing functions, such as swinging a golf club, that are best accomplished with automatic processes.See Dan Ariely et al. (2004) for the latest evidence.



21/56


quadrant I has a tendency to explain behav-ior in terms it can understand in terms of quadrant I processes.

3.4.1 Collaboration and CompetitionAlthough it is heuristically useful to distin-

guish between cognitive and affectiveprocesses, and between controlled and auto-matic processes, most judgments and behav-iors result from interactions between them.Collaboration, delegation of activity, andproper balance across the quadrants areessential to normal decision making. Many decision-making disorders may originate inan improper division of labor between the

quadrants. For example, psychiatry recog-nizes a decision-making continuum definedby the impulsive, light decision-makingstyle at one end and the compulsive, heavystyle at the other. The decisions of an impul-sive individual are excessively influenced by external stimuli, pressures, and demands.Such a person may not be able to give a moresatisfying explanation of an action exceptthat he felt like it (David Shapiro 1965). By contrast, an obsessivecompulsive person will subject even the most trivial decisions to

extensive deliberation and calculation. In sit-uations in which it is entirely appropriate tomake a quick decision based on impulsee.g., when choosing which video to rentfor the evening or what to order in arestaurantthe obsessive-compulsive willget stuck.

We are only now beginning to appreciatethe importance of affect for normal decisionmaking. The affective system provides inputsin the form of affective evaluations of behav-ioral optionswhat Damasio (1994) refers

to as somatic markers. Damasio and hiscolleagues show that individuals with mini-mal cognitive, but major affective deficitshave difficulty making decisions, and oftenmake poor decisions when they do (AntoineBechara et al. 1994; Bechara, Damasio,Damasio, and Lee 1999; Damasio 1994). It isnot enough to know what should be done;it is also necessary to feel it.

There is also interesting experimental evi-dence that deliberative thinking blocksaccess to ones emotional reactions to objects

and so reduces the quality of decisions (e.g., Wilson and Schooler 1991). In one study (Wilson et al. 1993) college students select-ed their favorite poster from a set of posters.Those who were instructed to think of rea-sons why they liked or disliked the postersbefore making their selection ended up lesshappy on average with their choice of posters (and less likely to keep them up ontheir dorm room wall) than subjects who were not asked to provide reasons.

Needless to say, affect can also distort cog-

nitive judgments. For example, emotionshave powerful effects on memorye.g., when people become sad, they tend to recallsad memories (which often increases theirsadness). Emotions also affect perceptions of risksanger makes people less threatenedby risks, and sadness makes them morethreatened (Jennifer Lerner and DacherKeltner 2001). Emotions also create moti- vated cognitionpeople are good at per-suading themselves that what they wouldlike to happen is what will happen. Quack

remedies for desperate sick people, and get-rich-quick scams are undoubtedly aided by the human propensity for wishful thinking. Wishful thinking may also explain high ratesof new business failure (Camerer and DanLovallo 1999), trading in financial markets,undersaving, and low rates of investment ineducation (foregoing large economicreturns). As LeDoux (1996) writes, Whileconscious control over emotions is weak,emotions can flood consciousness. This isso because the wiring of the brain at this

point in our evolutionary history is such thatconnections from the emotional systems tothe cognitive systems are stronger than con-nections from the cognitive systems to theemotional systems.

When it comes to spending money ordelaying gratification, taking or avoidingrisks, and behaving kindly or nastily towardother people, people often find themselves



22/56


of two minds; our affective systems drive usin one direction and cognitive deliberationsin another. We find ourselves almost compul-

sively eating our childrens left-overHalloween candy, while obsessing about howto lose the extra ten pounds; gambling reck-lessly at the casino, even as a small voice inour head tell us to rein it in; trying to build upcourage to step up to the podium; or tempt-ed to give a little to the pathetic street-cornerbeggar though we know our crumpled dollar will go further if donated to United Way.

All of these deviations occur because ouraffective system responds to different cues,and differently to the same cues, as our

cognitive system does. As Rolls (1999) writes,

emotions often seem very intense in humans,indeed sometimes so intense that they producebehaviour which does not seem to be adaptive,such as fainting instead of producing an activeescape response, or freezing instead of avoiding,or vacillating endlessly about emotional situa-tions and decisions, or falling hopelessly in loveeven when it can be predicted to be withouthope or to bring ruin. The puzzle is not only thatthe emotion is so intense, but also that even withour rational, reasoning, capacities, humans still

find themselves in these situations, and may findit difficult to produce reasonable and effectivebehaviour for resolving the situation (p. 282).

Such divergences between emotional reac-tions and cognitive evaluations arise, Rolls(1999) argues, because

in humans the reward and punishment systemsmay operate implicitly in comparable ways tothose in other animals. But in addition to this,humans have the explicit system (closely relatedto consciousness) which enables us consciously to look and predict many steps ahead. (p. 282).

Thus, for example, the sight or smell of acookie might initiate motivation to consumeon the part of the affective system, but mightalso remind the cognitive system that one ison a diet.

Exactly how cognitive and affective sys-tems interact in the control of behavior is not

well understood. At a neurological level, itappears that an organ in the reptilian braincalled the striatum (part of a larger system

called the basal ganglia) plays a critical role.The striatum receives inputs from all parts of the cerebral cortex, including the motor cor-tex, as well as from affective systems such asthe amygdala. Lesions of pathways that sup-ply dopamine to the striatum leads, in ani-mals, to a failure to orient to stimuli, a failureto initiate movements, and a failure to eat ordrink (J. F. Marshall et al. 1974). In humans,depletion of dopamine in the striatum isfound in Parkinsons disease, the most dra-matic symptom of which is a lack of volun-

tary movement. The striatum seems to beinvolved in the selection of behaviors fromcompetition between different cognitive andaffective systemsin producing one coher-ent stream of behavioral output, which canbe interrupted if a signal of higher priority isreceived, or a surprising stimulus appears(Carolyn Zink et al. 2003).

The extent of collaboration and competi-tion between cognitive and affective sys-tems, and the outcome of conflict when itoccurs, depends critically on the intensity of

affect (Loewenstein 1996; Loewenstein andLerner 2003). At low levels of intensity,affect appears to play a largely advisoryrole. A number of theories posit that emo-tions carry information that people use as aninput into the decisions they face (e.g.,Damasio 1994; Ellen Peters and Paul Slovic2000). The best-developed of theseapproaches is affect-as-information theory (Norbert Schwarz and Gerald Clore 1983;Schwarz 1990; Clore 1992).

At intermediate levels of intensity, peo-

ple begin to become conscious of conflictsbetween cognitive and affective inputs. Itis at such intermediate levels of intensity that one observes the types of efforts atself-control that have received so muchattention in the literature (Jon Elster 1977; Walter Mischel, Ebbe Ebbesen, andAntonette Zeiss 1972; Thomas Schelling1978, 1984).



23/56


Finally, at even greater levels of intensity,affect can be so powerful as to virtually pre-clude decision making. No one decides to

fall asleep at the wheel, but many people do.Under the influence of intense affectivemotivation, people often report themselves asbeing out of control or acting against theirown self-interest (Baumeister, Heatherton,and Tice 1994; Stephen Hoch andLoewenstein 1991; Loewenstein 1996). AsRita Carter writes, where thought conflicts with emotion, the latter is designed by theneural circuitry in our brains to win (1999).

3.4.2 Spurious Sense-Making

Sense-making is an important form of interaction between quadrant I and theother quadrants. The brains powerful drivetoward sense making leads us to strive tointerpret our own behavior. Since quadrant Ioften does not have conscious access toactivity in the other quadrants, it is perhapsnot surprising that it tends to over-attributebehavior to itselfi.e., to deliberative deci-sion processes. Even though much of thebrains activity is cognitively inaccessible, we have the illusion that we are able to make

sense of it, and we tend to make sense of itin terms of quadrant I processes.Research with EEG recordings has shown

(Benjamin Libet 1985) that the precisemoment at which we become aware of anintention to perform an action trails the ini-tial wave of brain activity associated with thataction (the EEG readiness potential) by about 300 msec. The overt behavioralresponse itself then follows the sensation of intention by another 200 msec. Hence, whatis registered in consciousness is a regular

pairing of the sensation of intention followedby the overt behavior. Because the neuralactivity antecedent to the intention is inac-cessible to consciousness, we experiencefree will (i.e., we cannot identify anythingthat is causing the feeling of intention).Because the behavior reliably follows theintention, we feel that this freely willedintention is causing the actionbut in fact,

17 the brain contains a specific cognitive system thatbinds intentional actions to their effects to construct acoherent experience of our own agency (Patrick Haggard,Sam Clark, and Jeri Kalogeras 2002).

both the sensation of intention and the overtaction are caused by prior neural events which are inaccessible to consciousness (see

Daniel Wegner and Thalia Wheatley 1999).17

Quadrant I tends to explain behavior ego-centricallyto attribute it to the types of deliberative processes that it is responsiblefor (see Richard Nisbett and Wilson 1977). Adramatic study demonstrating this phenom-enon was conducted with a split-brainpatient (who had an operation separating theconnection between the two hemispheres of his brain). The patients right hemispherecould interpret language but not speak, andthe left hemisphere could speak (LeDoux

1996). The patients right hemisphere wasinstructed to wave his hand (by showing the word wave on the left part of a visualscreen, which only the right hemisphereprocessed). The left hemisphere saw theright hand waving but was unaware of theinstructions that had been given to the righthemisphere (because the cross-hemisphereconnections were severed). When thepatient was asked why he waved, the lefthemisphere (acting as spokesperson for theentire body) invariably came up with a plau-

sible explanation, like I saw somebody Iknew and waved at them.

4. General Implications of Neuroscience for Economics

To add value to economics, neuroscienceneeds to suggest new insights and usefulperspectives on old problems. This sectiondiscusses some broad implications for eco-nomics of the ideas and findings reviewed inthe previous section. First, we show that

neuroscience findings raise questions aboutthe usefulness of some of the most commonconstructs that economists commonly use,such as risk aversion, time preference, and



24/56


altruism. Second, we show how the exis-tence of specialized systems challengesstandard assumptions about human infor-

mation processing and suggests that intelli-gence and its oppositeboundedrationalityare likely to be highly domain-specific. Third, brain-scans conducted whilepeople win or lose money suggest thatmoney activates similar reward areas as doother primary reinforcers like food anddrugs, which implies that money confers direct utility, rather than simply being val-ued only for what it can buy. Fourth, weshow that research on the motivational andpleasure systems of the brain human chal-

lenges the assumed connection betweenmotivation and pleasure. Finally, wedescribe some of the important implicationsof cognitive inaccessibility for economics.

4.1 Economic Constructs

Knowing how the brain solves problems,and what specialized systems it has at its dis-posal to do so, challenges some of our fun-damental assumptions about how peoplediffer from one-another when it comes toeconomic behavior. Economists currently

classify individuals on such dimensions astime preference, risk preference, andaltruism. These are seen as characteristicsthat are stable within an individual over timeand consistent across activities; someone who is risk-seeking in one domain is expect-ed to be risk-seeking in other domains as well. But empirical evidence shows that risk-taking, time discounting, and altruism are very weakly correlated or uncorrelatedacross situations. This inconsistency resultsin part from the fact that preferences are

state-contingent (and that people may notrecognize the state-contingency, whichif they didwould trigger overrides thatimpose more consistency than observed).But it also may point to fundamental prob-lems with the constructs that we use todefine how people differ from each other.

As an illustration, take the concept of time-preference. In empirical applications,

economic analyses typically assume that thesame degree of time preference is presentfor all intertemporal tradeoffs saving for

the future, flossing your teeth, dieting, andgetting a tattoo. For example, whether a per-son smokes is sometimes taken as a crudeproxy for low rates of time discounting, instudies of educational investment or savings.Thinking about the modularity of the brainsuggests that while different intertemporaltradeoffs may have some element of plan-ning in common (e.g., activity in prefrontalcortex), different types of intertemporalchoices are likely to invoke qualitatively dif-ferent mixtures of neural systems and hence

to produce entirely different patterns of behavior. Then measured discount rates willnot be perfectly correlated across domains,and might hardly be correlated at all.

Much as the study of memory has beenrefined by the identification of distinct mem-ory systems, each with its own properties of learning, forgetting, and retrieval; the study of intertemporal choice might be enhanced by asimilar decomposition, most likely informedby neuroscience research. For example,unpublished research by Loewenstein and

Roberto Weber suggests that, in normal sam-ples, future-oriented behaviors tend to clus-ter in tasks which tap different dimensions of self-control. For example, flossing your teethis statistically associated with a number of other minor, repetitive, helpful behaviorssuch as feeding parking meters religiously and being on time for appointments. Thesebehaviors seem to involve conscientious-ness, an important measure in personality theory.18 John Ameriks, Andrew Caplin, andJohn Leahy (in press) measured the propen-

sity to plan by asking TIAACREF enrolleesto disagree or agree with statements like Ihave spent a great deal of time developing afinancial plan. These measures correlate with actual savings rates.

Dieting and use of addictive drugs, incontrast to punctuality and flossing, mightbe involve very different circuitry andhence reveal fundamentally different dis-



25/56


18 Based on the observation that drugs affecting sero-tonin receptors are used to treat compulsion disorders,compulsivity appears to have some connection to the neu-rotransmitter serotonin, In addition, compulsion disordersare thought to be related to hyperactivity in the caudatenucleus, a reminder region of the limbic system.

counting behavior. Both of these activitiesinvolve visceral motives that seem to differreliably across persons and may be only

weakly linked to conscientiousness. If youre the kind of person who loves to eat,or drink alcohol, then resisting indulging inthose activities requires difficult exertion of self-control.

4.2 Domain-Specific Expertise

neuroeconomics - how neuroscience informs economics

Documents